TABLE OF CONTENTS

initialization/makedatamat [ Functions ]

NAME

    makedatamat --- make data matrix

FUNCTION

Convert an Anderson - Gill data frame for a single recurrent event process into the data matrix format used in the remainder of the algorithm.

SYNOPSIS

206 makedatamat <- function(agdata, as, alternating)

INPUTS

    agdata         a data frame in Anderson-Gill format with columns
                   i,j,k,r,start,stop,delta and covariates
    as             a matrix of discretization breakpoints, for each stratum
                   generated by makeas
    alternating    boolean, indicating whether the at-risk function 
                   alternates between events (episodic process)

OUTPUTS

    datamat        a matrix with columns i,j,k,r,time,delta,smin,smax
                   and covariates. The columns are:
                   i       cluster
                   j       subject
                   k       event counter
                   r       stratum
                   time    length of time for each interval
                   delta   event indicator
                   smin    discretization interval corresponding to
                           start time in input
                   smax    discretization interval corresponding to
                           stop time in input

SOURCE

209 {
210     # Allocate space. Most data is copied directly from agdata.
211     datamat <- as.matrix(cbind(agdata[, 1:4], 0, agdata$delta, 1, 0, 
212         agdata[, 8:dim(agdata)[2]]))
213     colnames(datamat)[5:8] <- c("time", "delta", "smin", "smax")
214     # Rows in which a new event has occurred
215     diffevent <- (c(TRUE, diff(agdata$i * 1000 + agdata$j) != 0) | 
216                   c(TRUE, agdata$delta[ - length(agdata$delta)] == 1))
217     badind <- NULL
218 
219     # Loop to compute discretization intervals
220     for(ind in 1:dim(datamat)[1])
221     {
222         # Reset last event start time
223         if(diffevent[ind]) {
224             lasteventtime <- agdata$start[ind]
225             smax <- 0
226         }
227        # Interevent time is given by stop - start from agdata.
228        if(!alternating){
229             newtime <- agdata$stop[ind] - lasteventtime
230         }else{
231             newtime <- agdata$stop[ind] - agdata$start[ind]
232         }
233         r <- datamat[ind, "r"]
234         datamat[ind, "time"] <- newtime
235         # Find the discretization intervals that the times fall into
236         smin <- smax + 1
237         smax <- sum(as[r, ] < newtime) 
238         # Most of the time, smin <= smax, however, if both start and stop times
239         # fall into the same interval, this needs to be fixed
240         if(smin > smax){
241             smin <- datamat[ind - 1, "smin"]
242             badind <- c(badind, ind - 1)
243             timediff0 <- as[r, datamat[ind - 1, "smax"] + 1] - 
244                          as[r, datamat[ind - 1, "smin"]]
245             timediff1 <- as[r, smax + 1] - as[r, smax]
246             datamat[ind, 9:dim(datamat)[2]] <- (timediff0 * 
247                 datamat[ind - 1, 9:dim(datamat)[2]] + timediff1 * 
248                 datamat[ind, 9:dim(datamat)[2]]) / (timediff0 + timediff1)
249         }
250         datamat[ind, "smin"] <- smin
251         datamat[ind, "smax"] <- smax
252 
253     }
254     # Remove bad entries
255     if(!is.null(badind)) datamat <- datamat[ - badind, ]
256     
257     return(datamat)
258 }